Data-Agnostic Augmentations for Unknown Variations: Out-of-Distribution Generalisation in MRI Segmentation

TL;DR

This paper explores data-agnostic augmentation methods like MixUp and Fourier Augmentation to improve the robustness and out-of-distribution generalization of MRI segmentation models in real-world clinical settings.

Contribution

It systematically evaluates alternative augmentation strategies that do not rely on specific distribution shift sources, enhancing model robustness in medical image segmentation.

Findings

Augmentation methods improve out-of-distribution generalization.

Enhanced feature separability and compactness.

Easy integration into existing training pipelines.

Abstract

Medical image segmentation models are often trained on curated datasets, leading to performance degradation when deployed in real-world clinical settings due to mismatches between training and test distributions. While data augmentation techniques are widely used to address these challenges, traditional visually consistent augmentation strategies lack the robustness needed for diverse real-world scenarios. In this work, we systematically evaluate alternative augmentation strategies, focusing on MixUp and Auxiliary Fourier Augmentation. These methods mitigate the effects of multiple variations without explicitly targeting specific sources of distribution shifts. We demonstrate how these techniques significantly improve out-of-distribution generalization and robustness to imaging variations across a wide range of transformations in cardiac cine MRI and prostate MRI segmentation. We quantitatively find that these augmentation methods enhance learned feature representations by promoting separability and compactness. Additionally, we highlight how their integration into nnU-Net training pipelines provides an easy-to-implement, effective solution for enhancing the reliability of medical segmentation models in real-world applications.